A neonatal intensive care unit environment early warning monitoring system and method

Abstract

The present application relates to the technical field of data analysis, in particular to a new-born baby monitoring room environment early warning monitoring system and method, comprising: a data acquisition module acquires environment information in a new-born baby monitoring room; a data processing module pre-processes the environment information; a data judgment module judges whether the pre-processed environment information belongs to a preset safety range, if not, generates early warning information; a data analysis module analyzes the environment information using a neural network model, identifies the cleaning state information of each medical staff's hand in the new-born baby monitoring room, judges whether the medical staff's hand is in a hand hygiene completion state, if not, generates hand cleaning prompt information; a central control early warning module receives the early warning information, and according to the early warning information, carries out early warning, receives the hand cleaning prompt information, and displays and pre-alerts. The present application can analyze the collected environment information in the new-born baby monitoring room in multiple directions, improve data utilization, and ensure the accuracy and comprehensiveness of monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a neonatal intensive care unit environment early warning monitoring system and method. Background Art

[0002] The Neonatal Intensive Care Unit (NICU) is a hospital department dedicated to the treatment and care of premature infants and other newborns requiring specialized medical care. Within the NICU, newborns are cared for by medical staff throughout their entire life. Maintaining stable temperature, humidity, sound, light intensity, and air quality within this environment is crucial for the health of newborns. Therefore, establishing an early warning and monitoring system for the NICU environment is essential.

[0003] Common early warning monitoring systems primarily monitor environmental parameters in neonatal intensive care units in real time, providing timely warnings when parameters exceed safety limits. They also support remote monitoring and management, allowing medical staff to stay informed of environmental conditions. However, these systems often analyze large amounts of collected data only within safety limits, resulting in a limited analysis method and low data utilization, hindering the accuracy and comprehensiveness of monitoring.

[0004] In particular, the early warning monitoring system only monitors the environmental parameters in the neonatal intensive care unit in real time. It does not monitor whether the medical staff who are the only ones in contact with the newborns comply with the contact requirements of the neonatal intensive care unit and accurately clean and disinfect their hands. Newborns in the neonatal intensive care unit are cared for by medical staff throughout the whole process. As the only outsiders who can come into contact with the newborns, medical staff need to comply with strict disinfection procedures and clean their hands to meet the standards when entering the neonatal intensive care unit to prevent the introduction of pathogens.

[0005] Therefore, there is an urgent need for a neonatal intensive care unit environmental early warning monitoring system and method, which can conduct multi-directional analysis of the collected environmental information in the neonatal intensive care unit, improve data utilization, ensure the accuracy and comprehensiveness of monitoring, and monitor and analyze whether medical staff comply with contact requirements based on the collected environmental information and perform accurate hand cleaning, thereby further ensuring the safety of the neonatal intensive care unit. Summary of the Invention

[0006] One of the purposes of the present invention is to provide a neonatal intensive care unit environment early warning monitoring system that can perform multi-directional analysis on the collected environmental information in the neonatal intensive care unit, improve data utilization, and ensure the accuracy and comprehensiveness of monitoring.

[0007] The present invention provides a basic solution 1: a neonatal intensive care unit environment early warning monitoring system, comprising:

[0008] Data acquisition module, used to collect environmental information in the neonatal intensive care unit;

[0009] A data processing module, used for pre-processing environmental information;

[0010] The data judgment module is used to determine whether the pre-processed environmental information falls within the preset safety range. If not, it generates an early warning message and sends it to the central control early warning module;

[0011] The data analysis module is used to analyze the pre-processed environmental information using a neural network model to identify the cleanliness status of the hands of medical staff in the neonatal intensive care unit. Based on the cleanliness status information, it determines whether the medical staff's hands are in a complete hand hygiene state. If not, it generates a hand cleaning reminder message and sends it to the central control warning module;

[0012] The central control warning module is used to receive warning information and issue warnings based on the warning information; it is also used to receive hand cleaning reminder information and display and pre-remind based on the hand cleaning reminder information.

[0013] Furthermore, the environmental information includes one or more of temperature and humidity information, air quality information, sound information, lighting information, and monitoring video image information.

[0014] Furthermore, the pre-processed environmental information is analyzed using a neural network model to identify the cleanliness status of the hands of each medical staff member in the neonatal intensive care unit, including:

[0015] Using the CNN neural network model, the monitoring video image information is used as input and the cleanliness status information of each medical staff member's hands is output;

[0016] The CNN neural network model includes: input layer, output layer, several convolutional layers, pooling layers, and fully connected layers;

[0017] The input layer inputs the monitoring video image information, which is then extracted through the convolution layer, compressed through the pooling layer, and the fully connected layer analyzes the cleanliness status of each medical staff member's hands based on the image features.

[0018] Furthermore, the specific structure of the CNN neural network model is: an input layer and an output layer, two convolutional layers, a pooling layer and a fully connected layer;

[0019] The pre-processed surveillance video image information is input through the input layer and transmitted to the first convolutional layer for image feature C1 extraction;

[0020] The image feature C1 is input to the first pooling layer for image feature compression to obtain the image feature S2. The strategy adopted by the pooling layer is maximum pooling.

[0021] Input the image feature S2 into the second convolutional layer to extract the image feature C3;

[0022] Input the image feature C3 into the second pooling layer for image feature compression to obtain image feature S4;

[0023] Input the image feature S4 into the first fully connected layer to analyze the cleanliness status of each medical staff member's hands, obtain the output result F5, and activate the output result F5 using the Leaky ReLU function;

[0024] The activated output result F5 is input to the second fully connected layer to analyze the cleanliness status information of each medical staff member's hands, obtain the output result F6, and activate the output result F6 using the Leaky ReLU function; the number of nodes in the first fully connected layer and the second fully connected layer are both greater than or equal to the number of image features S4;

[0025] The activated output result F6 is input into the output layer for conversion, and the final prediction result f(out) is output as the clean status information of the medical staff's hands.

[0026] Furthermore, historical surveillance video image information and medical staff hand hygiene status information were used to construct training and validation sets to train and validate the CNN neural network model, including:

[0027] The constructed CNN neural network model is trained using the training set;

[0028] The trained CNN neural network model is verified using the validation set. If it meets the preset requirements, the training is completed and the trained CNN neural network model is output;

[0029] If it does not meet the preset requirements, continue training the CNN neural network model;

[0030] During the training process, the weights of the nodes in the fully connected layer are updated using an optimization algorithm, including:

[0031] Randomly generate several weight schemes as individual schemes to form the initial population;

[0032] Construct a weight evaluation function based on the training evaluation value of the CNN neural network model;

[0033] According to the weight evaluation function, the weight scheme in the initial population is selected and iterative optimization is performed to obtain the optimal weight scheme. Furthermore, a weight evaluation function is constructed to evaluate the difference between the final predicted result f(out) and the actual result y, which is:

[0034] P=aE -1 ;

[0035] wherein P is a weight evaluation value, E is a mean square error loss function, and a is an adjustment constant;

[0036]

[0037] wherein f(out i ) is a final prediction result of the CNN neural network model output as input of the input data in the i-th set of training data in the training set, and y i is an actual result in the i-th set of training data in the training set.

[0038] Further, the weight scheme in the initial population is selected according to the weight evaluation function, and the optimal weight scheme is obtained by iterative optimization, comprising:

[0039] initializing the iteration number k = 1;

[0040] According to the individual scheme in the population, the weights in the CNN neural network model are set, and the same training set is used to train the CNN neural network model with different weights, and the corresponding final prediction result f(out i ) is obtained.

[0041] According to the mean square error loss function, the weight evaluation value P j of each individual scheme is calculated, wherein P j is the weight evaluation value of the j-th individual scheme.

[0042] According to the weight evaluation value P j of each individual scheme, a roulette wheel selection is performed in the initial population to select individual schemes; the selection probability is proportional to the weight evaluation value.

[0043] The selected individual schemes are subjected to mutation and crossover, and individual schemes that cannot be implemented are removed to form a new population; wherein the individual scheme that cannot be implemented is an individual scheme that cannot be implemented.

[0044] It is judged whether the current iteration number is equal to the preset number, if yes, the individual scheme with the maximum weight evaluation value is obtained as the optimal weight scheme, and if not, the iteration number k = k + 1 is updated, and the process of setting the weights in the CNN neural network model according to the individual scheme in the population, and training the CNN neural network model with different weights using the same training set to obtain the corresponding final prediction result f(out i ) is performed again.

[0045] Further, the weight evaluation value P j of each individual scheme is calculated according to the mean square error loss function, wherein P j is the weight evaluation value of the j-th individual scheme, and further comprising:

[0046] Compare the weight evaluation value of each individual solution in the k iteration to see if it is greater than the weight evaluation value of the corresponding parent in the k-1 iteration. If not, replace the individual solution in the k iteration with the individual solution of its parent. If so, execute the weight evaluation value P of each individual solution. j , using the roulette wheel method to select individual solutions from the initial population.

[0047] Furthermore, in the CNN neural network model, the image feature dimension is expanded according to the environmental information, the fully connected layer analyzes the cleanliness status information of each medical staff member's hands according to the image features of the feature dimension, and the output layer outputs the cleanliness status information of each medical staff member's hands.

[0048] Beneficial effects of this solution: This solution collects environmental information in the neonatal intensive care unit through the data acquisition module. In addition to the traditional method of determining whether the pre-processed environmental information falls within the preset safety range and, if not, generating an early warning message and sending it to the central control early warning module for early warning, this solution also sets up a data analysis module to analyze the pre-processed environmental information using a neural network model, identify the cleanliness status of the hands of medical staff in the neonatal intensive care unit, and determine whether the medical staff's hands are in a complete hand hygiene state based on the cleanliness status information. If not, a hand cleaning reminder message is generated, thereby increasing the analysis method and improving data utilization;

[0049] In particular, this solution constructs a neural network model to analyze surveillance video image information and uses historical data to train the neural network model to ensure the accuracy of the neural network model's predictions. At the same time, during the training of the neural network model, the weights of the nodes in the fully connected layer are updated using an optimization algorithm. The weight scheme is evaluated based on the constructed weight evaluation function, and cross-mutation and iterative optimization are performed to find the optimal weight scheme, thereby further improving the accuracy of the neural network model's predictions.

[0050] Multi-faceted early warning monitoring and analysis can effectively ensure the accuracy and comprehensiveness of neonatal intensive care unit monitoring.

[0051] In summary, this solution can conduct multi-directional analysis of the environmental information collected in the neonatal intensive care unit, improve data utilization, ensure the accuracy and comprehensiveness of monitoring, and monitor and analyze whether medical staff comply with contact requirements and perform accurate hand cleaning based on the collected environmental information, further ensuring the safety of the neonatal intensive care unit.

[0052] The second purpose of the present invention is to provide a neonatal intensive care unit environment early warning monitoring method, which can conduct multi-directional analysis of the collected environmental information in the neonatal intensive care unit, improve data utilization, ensure the accuracy and comprehensiveness of monitoring, and monitor and analyze whether medical staff comply with contact requirements based on the collected environmental information, perform accurate disinfection, and further ensure the safety of the neonatal intensive care unit.

[0053] The present invention provides a second basic solution: a method for early warning monitoring of a neonatal intensive care unit environment, which uses the above-mentioned early warning monitoring system for the neonatal intensive care unit environment, comprising:

[0054] Collect environmental information in the neonatal intensive care unit;

[0055] Preprocessing of environmental information;

[0056] Determine whether the pre-processed environmental information falls within the preset safety range. If not, generate warning information and issue an early warning based on the warning information;

[0057] The pre-processed environmental information is analyzed using a neural network model to identify the cleanliness status information of the hands of medical staff in the neonatal intensive care unit. Based on the cleanliness status information, it is determined whether the medical staff's hands are in a complete hand hygiene state. If not, a hand cleaning prompt information is generated and displayed and pre-reminders are given based on the hand cleaning prompt information. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The following is a logic block diagram of an embodiment of a neonatal intensive care unit environment early warning monitoring system of the present invention. DETAILED DESCRIPTION

[0059] The following is further described in detail through specific implementation methods:

[0060] Example 1

[0061] The embodiment is basically as shown in the attached Figure 1 Shown: A neonatal intensive care unit environment early warning monitoring system, comprising:

[0062] A data acquisition module is used to collect environmental information in the neonatal intensive care unit; the environmental information includes one or more of temperature and humidity information, air quality information, sound information, lighting information, and monitoring video image information;

[0063] The data processing module is used to pre-process the environmental information; the pre-processing includes data standardization and data cleaning; in particular, the pre-processing of the surveillance video image information includes: de-meaning, normalization, PCA dimensionality reduction and whitening;

[0064] The data judgment module is used to determine whether the pre-processed environmental information falls within the preset safety range. If not, it generates an early warning message and sends it to the central control early warning module;

[0065] Specifically, based on the temperature and humidity information, it is determined whether the temperature and humidity fall within the preset temperature and humidity range. If not, an early warning message containing temperature and humidity anomalies is generated and sent to the central control early warning module; wherein the temperature and humidity include temperature and humidity, and a preset temperature range and a preset humidity range are set accordingly. If the temperature or humidity does not fall within the corresponding preset temperature range or preset humidity range, an early warning message containing temperature and humidity anomalies is generated and sent to the central control early warning module;

[0066] Based on the air quality information, it is determined whether the air quality falls within the preset air quality range. If not, an alarm message containing air quality abnormality is generated and sent to the central control alarm module. The air quality information mainly includes: sulfur dioxide (SO2) concentration, nitrogen dioxide (NO2) concentration, inhalable particulate matter (PM10) concentration, carbon monoxide (CO) concentration, and fine particulate matter (PM2.5) concentration;

[0067] Based on the sound information, determine whether the sound falls within the preset sound range. If not, generate an alarm message containing the sound abnormality and send it to the central control alarm module;

[0068] Based on the lighting information, it is determined whether the lighting information meets the preset lighting intensity range of the current time node. If not, an early warning message containing lighting anomalies is generated and sent to the central control early warning module.

[0069] The data analysis module is used to analyze the pre-processed environmental information using a neural network model to identify the cleanliness status of the hands of medical staff in the neonatal intensive care unit. Based on the cleanliness status information, it determines whether the medical staff's hands are in a complete hand hygiene state. If not, it generates a hand cleaning reminder message and sends it to the central control warning module;

[0070] Specifically, the monitoring video image information includes: monitoring video images of medical staff washing their hands collected by a monitoring device set up in the hand washing area of ​​the neonatal intensive care unit; the neural network model evaluates the cleanliness status of the medical staff's hands based on the monitoring video image information of the medical staff washing their hands, and outputs an evaluation value, i.e., the cleanliness status information;

[0071] A CNN neural network model is used to take pre-processed surveillance video image information as input, output the cleanliness status information of each medical staff member's hands, and determine whether the cleanliness status information falls within the preset status value range. If so, the medical staff member's hands are determined to be in a complete hand hygiene state. If not, the medical staff member's hands are determined to be in an uncompleted hand hygiene state, and a hand cleaning prompt message is generated and sent to the central control warning module. The CNN neural network model determines whether the medical staff member's hand washing process complies with the preset hand washing actions, such as the seven-step hand washing method, based on the pre-processed surveillance video image information, and thus generates and outputs the cleanliness status information.

[0072] The CNN neural network model includes: input layer, output layer, several convolutional layers, pooling layers, and fully connected layers;

[0073] The input layer inputs the monitoring video image information, which is then passed through the convolution layer to extract image features, the pooling layer to compress image features, and the fully connected layer to analyze the cleanliness status of each medical staff member's hands based on the image features. The output layer outputs the cleanliness status information of the medical staff member's hands;

[0074] Specifically, the specific structure of the CNN neural network model in this embodiment is: an input layer and an output layer, two convolutional layers, a pooling layer and a fully connected layer;

[0075] The pre-processed surveillance video image information is input through the input layer and transmitted to the first convolutional layer for image feature C1 extraction;

[0076] The image feature C1 is input to the first pooling layer for image feature compression to obtain the image feature S2. The strategy adopted by the pooling layer is Max Pooling.

[0077] Input the image feature S2 into the second convolutional layer to extract the image feature C3;

[0078] Input the image feature C3 into the second pooling layer for image feature compression to obtain image feature S4;

[0079] Input the image feature S4 into the first fully connected layer to analyze the cleanliness status of each medical staff member's hands, obtain the output result F5, and activate the output result F5 using the Leaky ReLU function (non-linear mapping);

[0080] The activated output result F5 is input to the second fully connected layer to analyze the cleanliness status information of each medical staff member's hands, obtain the output result F6, and activate the output result F6 using the Leaky ReLU function; the number of nodes in the first fully connected layer and the second fully connected layer are both greater than or equal to the number of image features S4;

[0081] The activated output result F6 is input into the output layer (softmax layer) for conversion, and the final prediction result f(out) is output, which is the clean status information of the medical staff's hands.

[0082] In the process of building the CNN neural network model, historical surveillance video image information and medical staff's hand cleaning status information were used to construct the training set and validation set;

[0083] The constructed CNN neural network model is trained using the training set;

[0084] The trained CNN neural network model is verified using the validation set. If it meets the preset requirements, the training is completed and the trained CNN neural network model is output;

[0085] If it does not meet the preset requirements, continue training the CNN neural network model;

[0086] During the above training process, the weights of the nodes in the fully connected layer are updated using an optimization algorithm;

[0087] Specifically,

[0088] Randomly generate several weight schemes as individual schemes to form the initial population; specifically, the weight scheme is [w1,w2,…,w m ], where w m is the mth weight value;

[0089] Construct a weight evaluation function based on the training evaluation value of the CNN neural network model;

[0090] Specifically, a weight evaluation function is constructed to evaluate the difference between the final prediction result f(out) and the actual result y, which is:

[0091] P=aE -1 ;

[0092] Where P is the weight evaluation value, E is the mean square error loss function, and a is the adjustment constant;

[0093]

[0094] Among them, f(out i ) is the final prediction result output by the CNN neural network model of the input data in the i-th group of training data in the training set, y i is the actual result of the i-th group of training data in the training set;

[0095] According to the weight evaluation function, a weight scheme is selected from the initial population, and iterative optimization is performed to obtain the optimal weight scheme;

[0096] Specifically, the number of initialization iterations k=1;

[0097] According to the individual schemes in the population, the weights in the CNN neural network model are set, and the same training set is used to train the CNN neural network models with different weights to obtain the corresponding final prediction results f(out i );

[0098] According to the mean square error loss function, calculate the weight evaluation value P of each individual solution j , where is the weight evaluation value of the j-th individual solution;

[0099] According to the weight evaluation value P of each individual solution j , using the roulette wheel method to select individual solutions from the initial population; the selection probability is proportional to the size of the weight evaluation value;

[0100] The selected individual solutions are mutated and crossed to form a new population;

[0101] Determine whether the current number of iterations is equal to the preset number. If so, obtain the individual solution with the largest weight evaluation value as the optimal weight solution. If not, update the number of iterations k=k+1, and then execute the weights in the CNN neural network model according to the individual solutions in the population. Use the same training set to train the CNN neural network models with different weights to obtain the corresponding final prediction results f(out i ).

[0102] In other embodiments, the weight evaluation value P of each individual solution is j , using the roulette wheel method, before selecting individual solutions in the initial population, it also includes:

[0103] Determine whether there is an individual solution with a weight evaluation value greater than the preset weight evaluation value. If so, the individual solution with the largest weight evaluation value in the current population is taken as the optimal weight solution. If not, execute the solution based on the weight evaluation value P of each individual solution. j , a roulette wheel method is used to select individual solutions in the initial population, so that when a weight solution that meets the preset weight requirements is found, the iteration is ended early, which reduces the computing requirements and memory consumption and speeds up the training process.

[0104] The central control warning module is used to receive warning information and issue warnings based on the warning information; it is also used to receive hand cleaning reminder information and display and pre-reminders based on the hand cleaning reminder information;

[0105] Specifically, the central control warning module receives warning information containing different abnormalities and issues corresponding warnings. Specific warning methods include but are not limited to sound and light warnings, pop-up warnings, information push warnings, and other warning methods to ensure the effectiveness of warnings. The central control warning module is also used to receive hand cleaning reminders and display them accordingly, facilitating monitoring of the neonatal intensive care unit and ensuring the safety of newborns. A soft light reminder device is also installed in the handwashing area of ​​the neonatal intensive care unit. Based on the hand cleaning reminder, the soft light is controlled to illuminate, without affecting the internal lighting, while directly reminding medical staff that their hands are not properly cleaned.

[0106] In other embodiments, the data analysis module is also used to analyze using a neural network model to identify the cleaning status information of the hands of medical staff in the neonatal intensive care unit, and at the same time identify the cleaning time of the hands of the medical staff, which is also used as the cleaning status information. According to the cleaning status information, it is judged whether the hands of the medical staff are in a hand hygiene completion state. If not, a hand cleaning prompt information is generated and sent to the central control warning module; specifically, based on the collected monitoring video images of the medical staff washing their hands, a neural network model is used to identify the cleaning time from the start to the end of the user's hand washing, and it is judged whether the evaluation value is greater than the preset evaluation value, and the cleaning time is greater than the preset cleaning time. If so, it is determined that the hands of the medical staff are in a hand hygiene completion state; if not, it is determined that the hands of the medical staff are not in a hand hygiene completion state.

[0107] This embodiment further provides a neonatal intensive care unit environment early warning monitoring method, which uses the above-mentioned neonatal intensive care unit environment early warning monitoring system, including:

[0108] Collect environmental information in the neonatal intensive care unit;

[0109] Preprocessing of environmental information;

[0110] Determine whether the pre-processed environmental information falls within the preset safety range. If not, generate warning information and issue an early warning based on the warning information;

[0111] The pre-processed environmental information is analyzed using a neural network model to identify the cleanliness status information of the hands of medical staff in the neonatal intensive care unit. Based on the cleanliness status information, it is determined whether the medical staff's hands are in a complete hand hygiene state. If not, a hand cleaning prompt information is generated and displayed and pre-reminders are given based on the hand cleaning prompt information.

[0112] Example 2

[0113] This embodiment is basically the same as the above embodiment, except that: the weight evaluation value P of each individual solution is calculated according to the mean square error loss function. j , where is the weight evaluation value of the j-th individual solution, and also includes:

[0114] Compare the weight evaluation value of each individual solution in the k iteration to see if it is greater than the weight evaluation value of the corresponding parent in the k-1 iteration. If not, replace the individual solution in the k iteration with the individual solution of its parent. If so, execute the weight evaluation value P of each individual solution. j , a roulette wheel method is used to select individual solutions in the initial population, so as to retain better individual solutions, ensure that the priority is always in the optimal direction, improve the optimization effect, and ensure that the weight of the individual solution found can enable the CNN neural network model to have a better prediction effect.

[0115] Example 3

[0116] This embodiment is basically the same as the above embodiment, except that: in the CNN neural network model, the image feature dimension is expanded according to the environmental information, the fully connected layer analyzes the cleanliness status of each medical staff member's hands based on the image features of the feature dimension, and the output layer outputs the cleanliness status information of each medical staff member's hands;

[0117] Specifically, after obtaining the image feature S4, the temperature and humidity information, air quality information, and lighting information obtained at the same time as the input surveillance video image information are added to the feature vector of the image feature S4 as new row data to form a new multi-dimensional feature vector S4';

[0118] Input the feature vector S4' into the first fully connected layer to analyze the cleanliness status information of each medical staff member's hands, obtain the output result F5, and activate the output result F5 using the Leaky ReLU function (non-linear mapping);

[0119] The activated output result F5 is input to the second fully connected layer to analyze the cleanliness status information of each medical staff member's hands, obtain the output result F6, and activate the output result F6 using the Leaky ReLU function; the number of nodes in the first fully connected layer and the second fully connected layer are both greater than or equal to the number of image features S4;

[0120] The activated output result F6 is input into the output layer for conversion, and the final prediction result f(out) is output, which is the clean status information of the medical staff's hands.

[0121] In the CNN neural network model, when the image features output by the convolutional layer and pooling layer are used as input to the fully connected layer, the dimension of the feature vector is increased and combined with other environmental information, thereby improving the reference dimension of the prediction. This can comprehensively consider multiple factors, improve prediction accuracy, and further enhance the hand disinfection effect of medical staff.

[0122] Example 4

[0123] This embodiment is basically the same as the above embodiment, except that:

[0124] Hands should also be cleaned and disinfected when touching different newborns, because all newborns in the neonatal intensive care unit are sick newborns, and continuous contact with different newborns should be avoided to prevent cross infection;

[0125] Therefore, the data analysis module in this solution is also used to identify the action information of medical staff based on the pre-processed environmental information, and to determine whether there are any non-compliant actions or non-compliant action tendencies based on the action information. If so, a non-compliant prompt message is generated and sent to the central control warning module and the terminal of the medical staff who took the corresponding action as a reminder;

[0126] Specifically, the monitoring video image information also includes: full-area monitoring video images in the neonatal intensive care unit;

[0127] Label the surveillance video image information and divide it into video segments containing different action information;

[0128] Extract features from video clips, such as optical flow, 3D convolutional features (C3D), spatiotemporal feature maps, etc.

[0129] Construct a deep learning model and train the model, using the extracted features and their corresponding action information for training, so that it can learn the action pattern from the input features; in this embodiment, an RNN model is used;

[0130] Use the trained deep learning model to classify surveillance video image information and identify the action information that is taking place;

[0131] Construct a time series model and use the historical action sequence to construct input and output pairs. Train the time series model so that it can predict the next action based on the action information that is currently occurring and output the predicted next action information. In this embodiment, an LSTM model is used.

[0132] When medical staff enter the neonatal intensive care unit, the system analyzes whether they have performed hand cleaning actions within a preset time period. If not, it determines that there are non-compliant actions, generates non-compliant prompt information, and sends it to the central control warning module and the terminal of the medical staff who performed the corresponding action as a reminder; thus, the system monitors and reminds medical staff to clean their hands first after entering the neonatal intensive care unit;

[0133] When medical staff enter the neonatal intensive care unit, if the first prediction of the medical staff's next action information is contact with the newborn, it is queried whether the medical staff's action information within the preset time period before the current time point contains action information for hand cleaning. If not, it is determined that there is a tendency of non-compliant actions, and a non-compliant prompt message is generated and sent to the central control warning module and the terminal of the medical staff who took the corresponding action for reminder; the information includes hand cleaning action information, including hand washing in the hand washing area and hand cleaning with disinfectant, so as to supervise and remind medical staff to disinfect their hands after contacting the previous newborn and before contacting the next newborn to prevent cross infection.

[0134] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A neonatal intensive care unit environment early warning monitoring system, characterized in that: include: A data acquisition module is used to collect environmental information in the neonatal intensive care unit; the environmental information includes monitoring video image information; A data processing module, used for pre-processing environmental information; The data judgment module is used to determine whether the pre-processed environmental information falls within the preset safety range. If not, it generates an early warning message and sends it to the central control early warning module; The data analysis module is used to analyze the pre-processed environmental information using a neural network model to identify the cleanliness status of the hands of medical staff in the neonatal intensive care unit, including: Using the CNN neural network model, the monitoring video image information is used as input and the cleanliness status information of each medical staff member's hands is output; The CNN neural network model includes: input layer, output layer, several convolutional layers, pooling layers, and fully connected layers. The specific structure is: one input layer and output layer, two convolutional layers, pooling layers, and fully connected layers; The input layer inputs the monitoring video image information, which is then sequentially processed through the convolution layer for image feature extraction, the pooling layer for image feature compression, and the fully connected layer for analyzing the cleanliness status of each medical staff member's hands based on the image features. This includes: The pre-processed surveillance video image information is input through the input layer and transmitted to the first convolutional layer for image feature C1 extraction; The image feature C1 is input to the first pooling layer for image feature compression to obtain the image feature S2. The strategy adopted by the pooling layer is maximum pooling. Input the image feature S2 into the second convolutional layer to extract the image feature C3; Input the image feature C3 into the second pooling layer for image feature compression to obtain image feature S4; Input the image feature S4 into the first fully connected layer to analyze the cleanliness status of each medical staff member's hands, obtain the output result F5, and activate the output result F5 using the Leaky ReLU function; The activated output result F5 is input to the second fully connected layer to analyze the cleanliness status information of each medical staff member's hands, obtain the output result F6, and activate the output result F6 using the Leaky ReLU function; the number of nodes in the first fully connected layer and the second fully connected layer are both greater than or equal to the number of image features S4; The activated output result F6 is input into the output layer, converted, and the final prediction result is output. , as the cleanliness status information of medical staff’s hands; Using historical surveillance video image information and medical staff hand hygiene status information, we constructed training and validation sets to train and validate the CNN neural network model, including: The constructed CNN neural network model is trained using the training set; The trained CNN neural network model is verified using the validation set. If it meets the preset requirements, the training is completed and the trained CNN neural network model is output; If it does not meet the preset requirements, continue training the CNN neural network model; During the training process, the weights of the nodes in the fully connected layer are updated using an optimization algorithm, including: Randomly generate several weight schemes as individual schemes to form the initial population; According to the training evaluation value of the CNN neural network model, a weight evaluation function is constructed to evaluate the final prediction result and actual results The difference is: ； in is the weight evaluation value, is the mean square error loss function, is the adjustment constant; ； Among them, For the training set The final prediction result output by the CNN neural network model using the input data in the group training data as input, For the training set The actual results in the group training data; According to the weight evaluation function, a weight scheme is selected from the initial population and iterative optimization is performed to obtain the optimal weight scheme; including: Initialize the number of iterations k=1; According to the individual schemes in the population, the weights in the CNN neural network model are set, and the same training set is used to train the CNN neural network models with different weights to obtain the corresponding final prediction results. ; According to the mean square error loss function, the weight evaluation value of each individual solution is calculated , among which The weight evaluation value of each individual solution is also included: comparing whether the weight evaluation value of each individual solution in k iterations is greater than the weight evaluation value of the corresponding parent in k-1 iterations. If not, the individual solution in k iterations is replaced by the individual solution of its parent. If so, the weight evaluation value of each individual solution is executed. , using the roulette wheel method to select individual solutions from the initial population; the selection probability is proportional to the size of the weight evaluation value; The selected individual solutions are mutated and crossed, and the unfeasible individual solutions are eliminated to form a new population; Determine whether the current number of iterations is equal to the preset number. If so, obtain the individual solution with the largest weight evaluation value as the optimal weight solution. If not, update the number of iterations k=k+1, and then execute the individual solutions in the population to set the weights in the CNN neural network model. Use the same training set to train CNN neural network models with different weights to obtain the corresponding final prediction results. ; Based on the cleaning status information, determine whether the medical staff's hands are in a hand hygiene completion state. If not, generate a hand cleaning reminder message and send it to the central control warning module; The central control warning module is used to receive warning information and issue warnings based on the warning information; it is also used to receive hand cleaning reminder information and display and pre-remind based on the hand cleaning reminder information.

2. The neonatal intensive care unit environment early warning monitoring system according to claim 1, characterized in that: In the CNN neural network model, the image feature dimension is expanded according to the environmental information, the fully connected layer analyzes the cleanliness status information of each medical staff member's hands according to the image features of the feature dimension, and the output layer outputs the cleanliness status information of each medical staff member's hands.

Images